Data science can add value to any business that uses it well. From finding statistics and insights across workflows to hiring new candidates and helping staff make better-informed decisions, data science is valuable to any company in any industry. The biggest reason for its looming popularity is its ability to allow brands to communicate their story in an engaging and powerful manner. When brands and companies comprehensively utilize data, they can share their goal with their target audience, thereby creating better brand connections. After all, nothing connects with consumers like an effective and powerful story that can inculcate all human emotions. Precisely, Data science algorithms help perform complex data science tasks like prediction, classification, clustering, and others.

Paving the way for the success of businesses through its values, EXL was founded on the core values of innovation, collaboration, excellence, integrity, and respect. The team has the innate ability to work together with clients to improve business outcomes, operations, and customer experience. Through domain, analytics, and technology expertise, EXL serves as the indispensable partner for data-led businesses. Its cutting-edge analytics solutions are powered by domain focus, end-to-end capabilities, exceptional talent, and continuous innovation. The data scientists at EXL make sense of data and generate deeper insights that drive clients' businesses forward.

Varun Aggarwal, Vice President of Analytics at EXL, has been a part of the organization for 15 years. He has witnessed the exciting times of data analytics evolution and contributed through incessant innovations over these years.

An Accomplished Big Data Leader Deciphering Business Challenges 

Varun is deeply passionate about solving complex business problems for his clients with the application of advanced analytics. He loves to innovate in the field of predictive modeling, aimed at improving accuracy through better algorithms, increasing efficiency through automation of logic, and delivering benefits through actionable insights. Some of his work includes:

Hill Climbing Algorithm: This algorithm boosts model performance. It searches for the optimal subset of features from a given list, subject to a user-defined performance metric as its objective function. The procedure uses the hill-climbing iterative modeling process by evaluating all combinations of n features before climbing up to n+1.

Super Interactions: This algorithm captures non-linear relationships. It explores all possible n-way combinations of interactions. For n = 2 through 5, just 50 raw features can form over 2 million new variables! The procedure is suitably coupled with effective and efficient variable reduction techniques.

Segmentation Recommender: The segmentation recommender algorithm enables data segmentation decision-making. It evaluates a set of pre-defined strategic scenarios on given data and makes a recommendation for a single overall model or multiple-segmented models. The procedure blends business needs with statistical tests such as correlation sign flip, over-dependence on a specific predictor, and error pattern analysis.

Feature Clustering Enhancer: This algorithm selects predictive and representative features. It recommends variable selection based on joint analysis of unsupervised feature clustering outcome and supervised association analysis. The procedure provides flexibility to shortlist the top variables from each category.

Statistical Model Assessment and Review Tool (S.M.A.R.T.): This is a .Net and SQL-based analytics product that serves as a one-stop-shop solution for model monitoring. Key features include a dashboard with multi-level views, on-demand monitoring, scheduler, model governance, and fully automated model assessment.

Media Mix Modeling Optimizer: This is a statistical tool to optimize advertising mix with respect to sales revenue or marketing budget. This optimization capability provides actionable recommendations for spend allocation among various online and offline marketing channels.

Such analytics accelerators have translated into faster and improved outcomes at scale for Varun's clients, thereby creating "speed to value" differentiation and enabling better decisions for the data-led businesses.

Denting the Digital Space with Data Science Contributions  

Varun has contributed to the data analytics field not only in the capacity of an individual consultant but also by leading large teams comprising 200+ data scientists and by training 1000+ analytics professionals on predictive modeling. In fact, he designed and authored a comprehensive data science methodology training course that feeds into his organization's flagship training program for new hires.

Varun pioneered the research on machine learning at EXL a decade ago. He led a team of 20 data scientists to win the Heritage Health Prize competition, securing the second position among 1600+ participating teams globally. Over years, he co-authored several research papers and presented at international advanced data analysis, business analytics, and machine learning conferences facilitated by organizers such as North East SAS Users Group, New York Area SAS Users Group, IIM-Ahmedabad, and Analytics India Magazine. His research work spans data segmentation, feature engineering, feature selection, model training, and model validation.

Data Segmentation 

• Should I Build a Segmented Model? A Practitioner's Perspective, NYASUG Conference, January 14, 2010, Pace University, NY, US

Feature Engineering and Feature Selection 

• PROC LOGISTIC Plus: The Power of Variable Transformations, NESUG Conference, September 14-17, 2008, Pittsburgh, PA, US
• Feature Selection and Dimension Reduction Techniques in SAS, NESUG Conference September 11-14, 2011, Portland, ME, US
• Feature Engineering Strategies: A Practitioner's Guide, 5th IIMA International Conference on Advanced-Data Analysis, Business Analytics, and Intelligence, April 8-9, 2017, Ahmedabad, India

Model Training 

• Ensemble Hybrid Logit Model, KDD Cup 2010, Educational Data Mining Challenge, hosted by PSLC DataShop, July 2010
• Solving the CECL Riddle through Risk Analytics, 6th IIMA International Conference on Advanced Data Analysis, Business Analytics and Intelligence, April 6-7, 2019, Ahmedabad, India
• Credit Card Fraud Detection using Feature Engineering and Machine Learning, presented at Machine Learning Developers Summit 2022 organized by Analytics India Magazine and published by Association of Data Scientists, Lattice, The Machine Learning Journal, Volume-3, Issue-1, JanuaryMarch 2022

Model Validation 

• Retail Credit Risk Model Validation: Performance and Stability Aspects, 4th IIMA International Conference on Advanced Data Analysis, Business Analytics and Intelligence, April 11-12, 2015, Ahmedabad, India

In addition, Varun has co-authored a series of EXL white papers on credit loss forecasting.

Advanced Analytics at the Heart of Innovation 

Varun is an alumnus of Delhi School of Economics and joined EXL in 2007 through campus placement. With 15 years of experience in consulting and advanced analytics, he has solved complex problems for Fortune 500 companies across the banking, telecom, and utility sectors. These business problems include (but are not limited to) credit risk underwriting, customer churn management, credit loss forecasting, campaign response modeling, marketing mix optimization, cross-sell strategies, and customer segmentation. Currently, he is serving as Vice President of Analytics at EXL, leading a team of 200+ data scientists.

Varun is passionate about improving data science algorithms and is an ardent researcher in the fields of advanced analytics, machine learning, and deep learning with a focus on solving real business problems and delivering value to his clients. He has represented EXL at several data mining competitions and led the development of multiple analytics products. He has bagged more than 5 prestigious awards at EXL for innovation and intellectual capital development.